Intel Announces 14nm, 22nm Atom Chips Officially
xbitlabs.com
xbitlabs.com
For reference (from wikipedia):
2 nm — diameter of DNA helix
20 nm — width of bacterial flagellum
Another amazing number: if a chip is 2cm across and runs at 2 GHz then a light ray can go from one side to the other side of the chip about 10 times in a clock cycle. Note that light goes around the earth about 7 times per second. In other words: suppose you have quick child who takes one step of 20cm in each clock cycle. Then the child walks around the earth about 7 times per second.
This announcement sounds like this could finally happen in the next year or two. I look forward to the day when I can leave my battery charger at home.
Biggest battery drain I've found so far is (surprise, surprise) flash. Compiling is a hit but not an insane one, using Visual Studio on a WPF app.
The big difference with Intel's approach and ARM's approach: with ARM you can license the core and build your own SoC, which means there are many different SoC's to choose from for handset manufacturers. Being limited to just the chips Intel chooses to produce won't be very attractive, especially as you loose a very nice way of product differentiation. I wonder how handset manufacturers will work with that.
If there's going to be competition between ARM and Intel on this scale, they'll probably both have their own pro's: ARM is very customizable, up to the point where you can choose your own FPU next to all the other stuff you put on an SoC. Intel might have the advantage of power consumption and speed if they get their act together, and will be able to use smaller manufacturing processes earlier (11nm), but you'll have to use Intel's SoC, which so far don't exist, are unproven and probably will never offer the freedom ARM SoC's have.
Also Intel is still struggling to get just one Atom core to a low enough power consumption level. They can't improve the power consumption and performance in the same time. That means ARM chips will still be ahead for years in power consumption and performance (or performance/Watt).
From what I've noticed right now an Atom chip is like twice as expensive as a similar performance ARM chip. So it will be interesting to see if they will ever overcome that, too. Intel has never really liked making Atom chips because they make so much less money on them than on their other chips.That's why they haven't really improved them since 2008 when they launched Atom.
But they need to make SoC's, because ARM chips improve ~2.5x every 12 months, that's about 4x every 18 months, so twice as fast as x86 chips currently. That's why Intel is saying they want to start making SoC's and double Moore's Law. Because ARM is already doing that! But Intel won't have their 22nm 3D Atom until late 2013 or early 2014. The 32nm one won't be competitive. And I doubt the 22nm will be, too. It might be competitive in power consumption by then, but it won't be in performance, since ARM chips will keep the same low power consumption that they have now, but will increase performance 2.5x every year by then, while Intel can't do both in the same time.